Re: [Gretl-users] SUR systems
Am 27.05.2011 17:46, schrieb Javier García:
Hi, again:
Thanks, Sven, your suggestion works perfectly!!!
But I have another question about the methodology that is used. I have
been looking at Gretl User’s Guide, but I don’t find anything about SUR
systems.
Yes it's a known "issue" that the user guide is missing a chapter about
system estimation.
I’d rather know what technique uses Gretl to estimate SUR
systems. Maximum Likelihood? Feasible Generalized Least Squares?
ML is used when you select FIML in the drop-down menu. So it wouldn't
make much sense to have SUR also do ML, AFAICS. So yes, if I'm not very
mistaken then it's FGLS.
the menu we select iterative process (or something like that), does
it
mean that it uses time and again FGLS until the convergence is obtained?
You guessed right!
If this is the case, what matrix uses as covariance matrix, one of
“White’s type”?
The formula for the covariance matrix is given in the online help entry
for 'estimate' (which is the implicit command you are issuing when you
click ok in the system specification dialog). So in general it should be
the standard SUR approach w/o heteroscedasticity.
But it's a good question whether or in which cases robust covar
estimation applies to systems in gretl. In principle I think it does,
for example we recently had a discussion about how to handle VARs with
robust standard errors. But I've just noticed there is no option in the
system specification dialog. You could however experiment a little bit:
in the preferences, you can activate the option "use HAC by default" or
something like that. And you can change the type of robust estimator.
Then you can re-estimate your system and check if the reported standard
errors have changed.
hth,
sven